@holmes-lab/holmes-kit 0.3.5 → 0.3.6

package/CHANGELOG.md

@@ -4,6 +4,33 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+<!-- @implements A-SPEC-209 -->
+## [0.3.6] - 2026-09-01
+
+Traceability hardening: comma-listed anchors stop silently losing ids, the claim gate stops
+false-flagging test fixtures, and boilerplate acceptance criteria can no longer be sealed.
+
+### Fixed
+
+- **Comma-listed anchors** (A-SPEC-503.1, field-measured +410 lost rtm edges): all set-semantics
+  anchor consumers — coverage (`extractAnchors`), graph (`implementsSpecs`/wrong-kind warning) and
+  the claim gate — now parse the full `@implements ID, ID, …` list through ONE shared parser. A
+  wrong-kind id in second position is finally visible to the warning; prose mentions after the
+  list stay non-anchors; single-id behavior is value-identical.
+- **Claim-gate preprocessing asymmetry** (A-SPEC-504.1): the gate judged raw text while the
+  scanner stripped string literals, so anchor-shaped FIXTURE strings were refused as unapproved
+  claims (measured twice) — breeding a string-concatenation evasion habit. Both surfaces now share
+  `stripStringLiterals`; a string-anchor is a claim to nobody, exactly as it is an anchor to
+  nobody (no protection lost — pinned as a test).
+
+### Added
+
+- **Acceptance-substance sealing gate** (A-SPEC-505.1, dogfooded: 18 boilerplate-criteria REQs
+  sealed in one day): `spec_approve` now refuses a REQ whose Success Criteria are absent or
+  entirely boilerplate, with cause and remedy in the refusal. Act-time placement — the 154/428
+  legacy approved REQs measured in this repo are untouched. The `spec_slice_init` REQ template no
+  longer generates that very boilerplate (empty Success Criteria scaffold: fill before sealing).
+
 <!-- @implements A-SPEC-209 -->
 ## [0.3.5] - 2026-09-01
 

package/dist/.build-id

@@ -1 +1 @@
-f98587b-mthg333w
+c3e5866-mthi2ccn

package/dist/holmes/cpg/cpg-scanner.js

@@ -106,12 +106,8 @@ function isVendorDir(dir) {
         return false;
     }
 }
-const IMPL = /@implements\s+(A-SPEC-\d{3,}(?:\.\d+)?)/g;
-// Well-formed governed-but-non-A-SPEC ids used (wrongly) as an @implements
-// anchor. Deliberately requires the full "KIND-NNN" shape (kind + '-' +
-// 3+ digits) so prose/comments like `@implements A-SPEC refs` (no id number)
-// or a bare word never match — only a genuine wrong-kind spec-id does.
-const WRONG_KIND_IMPL = /@implements\s+((?:REQ|H-SPEC|C-SPEC|T-SPEC)-\d{3,}(?:\.\d+)?)/g;
+// Anchor parsing (incl. wrong-kind detection) moved to the shared comma-list parser in
+// rtm/anchor-ids (A-SPEC-503.1) — the local single-capture regexes dropped every id after a comma.
 /** Single source of truth for which file extensions CpgScanner ingests (REQ-124 gate 2). */
 exports.SCANNABLE_EXTENSIONS = ['.ts', '.mts', '.cts', '.tsx', '.jsx', '.js', '.mjs', '.cjs', '.py', '.cs', '.java', '.go', '.rs', '.cpp', '.cc', '.cxx', '.hpp', '.hh', '.h'];
 const SCANNABLE_RE = /\.(ts|mts|cts|tsx|jsx|js|mjs|cjs|py|cs|java|go|rs|cpp|cc|cxx|hpp|hh|h)$/;
@@ -122,6 +118,7 @@ var test_files_1 = require("./test-files");
 Object.defineProperty(exports, "TEST_FILE_PATTERNS", { enumerable: true, get: function () { return test_files_1.TEST_FILE_PATTERNS; } });
 Object.defineProperty(exports, "isTestFile", { enumerable: true, get: function () { return test_files_1.isTestFile; } });
 const test_files_2 = require("./test-files");
+const anchor_ids_1 = require("../rtm/anchor-ids");
 // .ts/.mts/.cts can contain TS type-assertion syntax (`<T>x`, arrow-generics)
 // that the 'tsx' grammar misreads as JSX, so they must stay on 'typescript'.
 // Everything else in SCANNABLE_EXTENSIONS (.tsx/.jsx/.js/.mjs/.cjs) never uses
@@ -212,8 +209,11 @@ class CpgScanner {
                         // symbol walk, and disambiguating a bare `.h` between C and C++
                         // is out of scope for this slice.
                         const lang = /\.py$/.test(e.name) ? 'python' : (/\.cs$/.test(e.name) ? 'csharp' : (/\.java$/.test(e.name) ? 'java' : (/\.go$/.test(e.name) ? 'go' : (/\.rs$/.test(e.name) ? 'rust' : (/\.(cpp|cc|cxx|hpp|hh|h)$/.test(e.name) ? 'cpp' : (TSX_GRAMMAR_RE.test(e.name) ? 'tsx' : 'typescript'))))));
-                        const implementsSpecs = [...code.matchAll(IMPL)].map(m => m[1]);
-                        const unanchoredImplements = [...code.matchAll(WRONG_KIND_IMPL)].map(m => m[1]);
+                        // @implements A-SPEC-503.1 — comma-listed anchors: the old single-capture regexes
+                        // dropped every id after the first (C12), and a wrong-kind id in second position was
+                        // invisible to the warning. Both consumers now ride the shared list parser.
+                        const implementsSpecs = (0, anchor_ids_1.anchorSpecIds)(code);
+                        const unanchoredImplements = (0, anchor_ids_1.wrongKindAnchorIds)(code);
                         const sourcePath = (0, source_path_1.toSourcePath)(repoRoot, path.resolve(p));
                         // @implements A-SPEC-140.1
                         // The third parse happens only when asked for. `undefined` from the parser (a language

package/dist/holmes/guardrail/anchors.js

@@ -15,9 +15,19 @@ exports.newlyClaimed = newlyClaimed;
  * keeps LLM judgement out of governance. What a change CLAIMS, though, is right there in the text —
  * that is the decidable subset, and this is it.
  */
-const ANCHOR_RE = /@implements\s+([A-Za-z]+-SPEC-[\w.]+)/g;
+// @implements A-SPEC-503.1 — comma-list support with the gate's OWN lenient token grammar kept
+// (any `*-SPEC-*` shape claims; the shared parser's strict kinds must not narrow a gate). The old
+// single capture let the second id of a listed claim slip past `newlyClaimed` unjudged (C12).
+const anchor_ids_1 = require("../rtm/anchor-ids");
+const ANCHOR_TOKEN = String.raw `[A-Za-z]+-SPEC-[\w.]+`;
+const ANCHOR_RE = new RegExp(String.raw `@implements\s+(${ANCHOR_TOKEN}(?:[ \t]*,[ \t]*${ANCHOR_TOKEN})*)`, 'g');
 function claimedAnchors(content) {
-    return [...new Set([...content.matchAll(ANCHOR_RE)].map((m) => m[1]))];
+    // @implements A-SPEC-504.1 — judge on the SAME literal-stripped text the coverage scanner uses:
+    // a string-anchor is an anchor to nobody, so treating it as a claim produced only false-positive
+    // refusals (measured twice on legitimate fixtures) and bred the '@'+'implements' evasion habit.
+    // Losing it costs no protection — there is no "claimed but unjudged" state to protect against.
+    const ids = [...(0, anchor_ids_1.stripStringLiterals)(content).matchAll(ANCHOR_RE)].flatMap((m) => m[1].split(/[ \t]*,[ \t]*/));
+    return [...new Set(ids)];
 }
 /**
  * The anchors this write ADDS, relative to what the file already had.

package/dist/holmes/mcp/handlers.js

@@ -1295,6 +1295,16 @@ function makeRawHandlers(store, opts) {
             if (stubs.length > 0) {
                 return { ok: false, reason: (0, approval_blockers_1.placeholderMessage)(stubs) };
             }
+            // @implements A-SPEC-505.1 — acceptance substance, judged at the ACT on the same candidate
+            // the seal would freeze, like the placeholder gate above (154/428 approved REQs here are
+            // non-stated legacy; a validateSpec predicate would brick them all). Judged HERE and not
+            // before the approval channels: an earlier extra read shifts the optimistic-concurrency
+            // window and lets a mid-approval edit get sealed (store-integrity contract: edits win,
+            // approvals lose). The post-grant refusal it costs is the standing property of every
+            // act-time blocker in this block (breaking_change, placeholder) — one ordering, one truth.
+            const unactionable = (0, approval_blockers_1.unactionableCriteriaBlocker)(candidate);
+            if (unactionable)
+                return { ok: false, reason: unactionable };
             // Parents-first: a sealed child snapshotting an unsealed parent would pin nothing. This loop
             // exists to COLLECT the digests; the refusal inside it is now a backstop, because
             // `parentBlockers` above already returns for the same condition with the same sentence. Kept
@@ -2841,7 +2851,6 @@ ${a.objective}
 - Standard project governance rules.
 
 ## Success Criteria
-- Implementation completed and verified by tests.
 
 ## Out of Scope
 - Unrelated feature changes.

package/dist/holmes/rtm/anchor-ids.d.ts

@@ -0,0 +1,15 @@
+/**
+ * The ONE comma-list grammar for anchor lines — C12, dogfooded: four parsers each captured only
+ * the first id of a comma-listed anchor (`… <marker> A-SPEC-…110, …219`), so every id after the
+ * comma silently vanished from coverage, graph edges (+410 recovered in the field by
+ * hand-normalizing 14 files) and claim gating. The list ends at the comma chain: a prose mention
+ * after the ids (`… (see another id)`) is NOT a claim, which is what keeps this widening from
+ * turning commentary into governance.
+ */
+export declare function stripStringLiterals(text: string): string;
+/** Every anchor-marker occurrence's comma-list, as an array of ids per occurrence. */
+export declare function implementsIdLists(text: string): string[][];
+/** Flattened A-SPEC ids across all lists — the graph/coverage consumers' shape. */
+export declare function anchorSpecIds(text: string): string[];
+/** Flattened wrong-kind ids (a governed id anchored where only A-SPEC belongs) — anywhere in a list. */
+export declare function wrongKindAnchorIds(text: string): string[];

package/dist/holmes/rtm/anchor-ids.js

@@ -0,0 +1,39 @@
+"use strict";
+// @implements A-SPEC-503.1
+/**
+ * The ONE comma-list grammar for anchor lines — C12, dogfooded: four parsers each captured only
+ * the first id of a comma-listed anchor (`… <marker> A-SPEC-…110, …219`), so every id after the
+ * comma silently vanished from coverage, graph edges (+410 recovered in the field by
+ * hand-normalizing 14 files) and claim gating. The list ends at the comma chain: a prose mention
+ * after the ids (`… (see another id)`) is NOT a claim, which is what keeps this widening from
+ * turning commentary into governance.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.stripStringLiterals = stripStringLiterals;
+exports.implementsIdLists = implementsIdLists;
+exports.anchorSpecIds = anchorSpecIds;
+exports.wrongKindAnchorIds = wrongKindAnchorIds;
+// @implements A-SPEC-504.1
+// The ONE string-literal preprocessing every anchor-judging surface shares. Measured 2026-09-01:
+// the coverage scanner stripped literals while the claim gate judged raw text, so a legitimate
+// fixture string was refused twice as an "unapproved claim" — and that friction bred the
+// '@'+'implements' evasion habit. A string-anchor is an anchor to nobody and must be a claim to
+// nobody; one definition here keeps the two judgments from drifting apart again.
+const STRINGS_RE = /'(?:\\.|[^'\\\n])*'|"(?:\\.|[^"\\\n])*"|`(?:\\.|[^`\\])*`/g;
+function stripStringLiterals(text) {
+    return text.replace(STRINGS_RE, '""');
+}
+const ID = String.raw `(?:REQ|A-SPEC|H-SPEC|C-SPEC|T-SPEC)-\d{3,}(?:\.\d+)?`;
+const LIST_RE = new RegExp(String.raw `@implements[ \t]+(${ID}(?:[ \t]*,[ \t]*${ID})*)`, 'g');
+/** Every anchor-marker occurrence's comma-list, as an array of ids per occurrence. */
+function implementsIdLists(text) {
+    return [...text.matchAll(LIST_RE)].map((m) => m[1].split(/[ \t]*,[ \t]*/));
+}
+/** Flattened A-SPEC ids across all lists — the graph/coverage consumers' shape. */
+function anchorSpecIds(text) {
+    return [...new Set(implementsIdLists(text).flat().filter((id) => id.startsWith('A-SPEC-')))];
+}
+/** Flattened wrong-kind ids (a governed id anchored where only A-SPEC belongs) — anywhere in a list. */
+function wrongKindAnchorIds(text) {
+    return [...new Set(implementsIdLists(text).flat().filter((id) => !id.startsWith('A-SPEC-')))];
+}

package/dist/holmes/rtm/test-scope.js

@@ -41,6 +41,7 @@ exports.extractAnchors = extractAnchors;
 exports.scanTestAnchors = scanTestAnchors;
 exports.computeTestScope = computeTestScope;
 // @implements A-SPEC-121.2
+const anchor_ids_1 = require("./anchor-ids");
 const fs = __importStar(require("node:fs"));
 const path = __importStar(require("node:path"));
 const spec_types_1 = require("../spec/spec-types");
@@ -256,12 +257,22 @@ function countTestCases(source, filePath) {
 // every id merely MENTIONED in it — a live run credited 50 cases to the nonexistent A-SPEC-999 from
 // a fixture string, and injecting one comment into any test file re-credited it to an arbitrary
 // spec). Both the fixture-string and the mid-code-line injection vectors die here.
-const IMPL_LINE_RE = /^[ \t]*(?:\/\/|\/\*|\*|#)[^\n]*?@implements\s+(A-SPEC-\d{3,}(?:\.\d+)?)/gm;
-const STRINGS_RE = /'(?:\\.|[^'\\\n])*'|"(?:\\.|[^"\\\n])*"|`(?:\\.|[^`\\])*`/g;
+// @implements A-SPEC-503.1 — the line must still BE a comment line, but the ids on it come from
+// the shared comma-list parser: the old single-capture regex silently dropped every id after the
+// first comma (C12 — +410 rtm edges recovered in the field only by hand-normalizing 14 files).
+const IMPL_LINE_RE = /^[ \t]*(?:\/\/|\/\*|\*|#)[^\n]*?@implements[ \t]/gm;
 /** A-SPEC ids anchored by a file, counting only standalone comment-line markers outside literals. */
 function extractAnchors(source) {
-    const noStrings = source.replace(STRINGS_RE, '""');
-    return [...new Set([...noStrings.matchAll(IMPL_LINE_RE)].map((m) => m[1]))];
+    // @implements A-SPEC-504.1 — the literal-stripping moved to the shared definition the claim
+    // gate now uses too; a private copy here is how the two judgments drifted apart in the first place.
+    const noStrings = (0, anchor_ids_1.stripStringLiterals)(source);
+    const ids = [];
+    for (const line of noStrings.split('\n')) {
+        IMPL_LINE_RE.lastIndex = 0;
+        if (IMPL_LINE_RE.test(line))
+            ids.push(...(0, anchor_ids_1.anchorSpecIds)(line));
+    }
+    return [...new Set(ids)];
 }
 /**
  * @implements A-SPEC-269

package/dist/holmes/spec/approval-blockers.d.ts

@@ -76,3 +76,13 @@ export declare const blockerClause: (label: string, blockers: string[]) => strin
  * Never throws: a refusal that says less is recoverable, a hook that dies is not.
  */
 export declare function blockerSummary(spec: Spec | undefined, resolve: (id: string) => Spec | null, specId?: string): string | null;
+/**
+ * The acceptance-substance gate, standing at the SEALING ACT like the placeholder gate above and
+ * for that gate's measured reason (A-SPEC-182): a static predicate in validateSpec would turn the
+ * already-approved corpus into violations — measured 2026-09-01, 154/428 approved REQs here are
+ * non-stated legacy — and brick the harness. Dogfooded origin: 18 REQs sealed in one day whose
+ * entire Success Criteria read "implementation completed and verified by tests", while the
+ * detector (acceptanceQuality) sat wired only into the post-hoc analyzer. The detector is
+ * CONSUMED, not copied — one judgment, two surfaces.
+ */
+export declare function unactionableCriteriaBlocker(spec: Spec): string | null;

package/dist/holmes/spec/approval-blockers.js

@@ -5,6 +5,9 @@ exports.placeholderSections = placeholderSections;
 exports.parentBlockers = parentBlockers;
 exports.approvalBlockers = approvalBlockers;
 exports.blockerSummary = blockerSummary;
+exports.unactionableCriteriaBlocker = unactionableCriteriaBlocker;
+// @implements A-SPEC-182
+const acceptance_quality_1 = require("./acceptance-quality");
 const validator_1 = require("./validator");
 const breaking_change_1 = require("./breaking-change");
 const spec_digest_1 = require("./spec-digest");
@@ -245,3 +248,23 @@ function blockerSummary(spec, resolve, specId) {
         return null;
     }
 }
+// @implements A-SPEC-505.1
+/**
+ * The acceptance-substance gate, standing at the SEALING ACT like the placeholder gate above and
+ * for that gate's measured reason (A-SPEC-182): a static predicate in validateSpec would turn the
+ * already-approved corpus into violations — measured 2026-09-01, 154/428 approved REQs here are
+ * non-stated legacy — and brick the harness. Dogfooded origin: 18 REQs sealed in one day whose
+ * entire Success Criteria read "implementation completed and verified by tests", while the
+ * detector (acceptanceQuality) sat wired only into the post-hoc analyzer. The detector is
+ * CONSUMED, not copied — one judgment, two surfaces.
+ */
+function unactionableCriteriaBlocker(spec) {
+    if (spec.type !== 'REQ')
+        return null;
+    const quality = (0, acceptance_quality_1.acceptanceQuality)(spec);
+    if (quality === 'stated')
+        return null;
+    return quality === 'absent'
+        ? `REQ ${spec.id}의 Success Criteria가 비어 있습니다 — 관측 가능한 기준을 채운 뒤 승인하십시오`
+        : `REQ ${spec.id}의 Success Criteria가 전량 보일러플레이트입니다('구현·테스트 완료'는 충족 정의가 아닙니다) — 각 항목이 검증 동사·측정 대상을 갖게 고친 뒤 승인하십시오`;
+}

package/package.json

@@ -1,7 +1,7 @@
 {
   "//": "@implements A-SPEC-209",
   "name": "@holmes-lab/holmes-kit",
-  "version": "0.3.5",
+  "version": "0.3.6",
   "description": "Holmes-Kit — deterministic Agentic Software Engineering (ASE) harness with causal traceability (spec chain + D-CPG + RTM + phase guardrail)",
   "main": "dist/holmes/mcp/server.js",
   "types": "dist/holmes/mcp/server.d.ts",